Generative AI Engineer
· About Me
Anubhab
Das
I'm a Generative AI Engineer who builds and ships LLM-powered systems in production. Currently a Software Developer at AIVOA.ai, building an agentic eQMS platform for pharmaceutical / GxP clients — where accuracy, traceability and regulatory compliance aren't optional. I work across agentic pipelines, RAG, workflow orchestration and the full-stack product around them.
- Agentic AI with LangGraph & MCP
- RAG Systems & Vector Search (Pinecone)
- LLM Evaluation & Self-Critique Loops
- Workflow Orchestration with Temporal
- FastAPI, PostgreSQL & React
· Expertise
Core Skills
LLM Applications
Production LLM chatbots and assistants with prompt engineering, tool-calling and structured outputs, integrated into enterprise apps through APIs.
RAG Systems
Retrieval-augmented generation with Pinecone — query expansion, cross-encoder reranking and data pipelines that feed grounded context.
Agentic AI
Multi-step agents with LangGraph and MCP — autonomous browsing agents (Browser Use, Playwright) and dynamic tool selection.
Workflow Orchestration
Durable, multi-stage LLM workflows with Temporal — including iterative self-critique loops for investigation analysis.
LLM Evaluation & Compliance
Rubric-based response evaluation and traceability for regulated (pharma / GxP) domains where accuracy is non-negotiable.
Backend & Full-Stack
FastAPI, PostgreSQL (recursive CTEs), SQLAlchemy, React with Redux / RTK Query and React Flow, plus PDF reporting with Jinja2 / WeasyPrint.
· Experience
Career Journey
Software Developer (Intern)
Bangalore, India · Remote
Building an agentic, AI-powered eQMS (electronic Quality Management System) for pharmaceutical / GxP clients, where output accuracy, traceability and regulatory compliance are core requirements.
- Developed a 5-stage Temporal workflow orchestrating LLM-powered investigation analysis for the Non-Conformance module, with an iterative self-critique loop that scores each response against a 10-point quality rubric.
- Built an autonomous documentation agent (Browser Use, Playwright, LangGraph) that crawls the platform and structures content into a navigation knowledge base feeding a Pinecone-backed RAG pipeline.
- Designed a context-isolated session architecture for the AI chat assistant, enabling multi-record conversational workflows across QMS record types.
- Built a Python calculation engine for cleaning-validation limits (Maximum Allowable Carryover) — HBEL/PDE, dose-based, 10ppm and LD50/NOEL methods per FDA/EMA/PIC/S/WHO — plus a worst-case-product ranking engine.
- Built a recursive PostgreSQL CTE-based traceability graph with React Flow/Dagre visualization and role-based access control, replacing a slower Python-level traversal.
- Architected a 9-stage Supplier Corrective Action Request (SCAR) workflow end-to-end, including seed datasets and a Jinja2/WeasyPrint HTML-to-PDF report pipeline.
· Background
Education & Certifications
Education
B.Tech in Computer Science
Elitte College of Engineering
2023 — 2026 · CGPA 8 / 10
Certifications
- Web Development — DTH/Internship/2024/86